Future Not Far releases self-evolving world action model WAM

Future Not Far released the Self-Evolving WAM self-evolving world action model, integrating joint world-action modeling, interactive experience learning, and training-effect feedback into a single system. According to its published data from update attempts in rounds 0 through 30, the autonomous full-completion success rates for four household tasks all improved: folding clothes rose from 54.6% to 96.8%, hanging clothes from 46.2% to 91.2%, making pasta from 42.7% to 89.5%, and stir-fried pork with chili peppers from 38.2% to 68.9%. These figures are counted as "autonomously completed from start to finish": completions after human takeover are not counted as autonomous successes, while timeouts and failures are still counted. The model is trained on 10,000 hours of its own first-person video and 30,000 hours of Futuring Robot F1/F2 interaction data. Future Not Far said the model will also be deployed on its upcoming third-generation robot, bringing self-evolution capability to a platform costing in the thousand-yuan range.

AI 根据报道生成 · 7 days ago更新